nvme: New function to perform a NVMe subsystem reset

This commit introduces spdk_nvme_ctrlr_subsystem_reset to
perform a NVMe subsystem reset according to the NVMe spec.

Signed-off-by: Michael Haeuptle <michael.haeuptle@hpe.com>
Change-Id: If4ffae1bd92d4d16a62ec2b6a01f7373223b5705
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5488
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Vasuki Manikarnike <vasuki.manikarnike@hpe.com>
Community-CI: Broadcom CI
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>

CHANGELOG.md

@@ -24,6 +24,10 @@ An `opts_size` element was added in the `spdk_blob_open_opts` structure to solve
 ABI compatiblity issue between different SPDK version. And also add `opts_size`
 parameter in `spdk_blob_open_opts_init` function.
 
+### nvme
+
+Added a new function `spdk_nvme_ctrlr_reset_subsystem` to perform a NVMe
+subsystem reset. Note: The NVMf target does not support the subsystem reset yet.
 
 ### event
 

include/spdk/nvme.h

@@ -900,6 +900,25 @@ void spdk_nvme_ctrlr_set_remove_cb(struct spdk_nvme_ctrlr *ctrlr,
  */
 int spdk_nvme_ctrlr_reset(struct spdk_nvme_ctrlr *ctrlr);
 
+/**
+ * Perform a NVMe subsystem reset.
+ *
+ * This function should be called from a single thread while no other threads
+ * are actively using the NVMe device.
+ * A subsystem reset is typically seen by the OS as a hot remove, followed by a
+ * hot add event.
+ *
+ * Any pointers returned from spdk_nvme_ctrlr_get_ns(), spdk_nvme_ns_get_data(),
+ * spdk_nvme_zns_ns_get_data(), and spdk_nvme_zns_ctrlr_get_data()
+ * may be invalidated by calling this function. The number of namespaces as returned
+ * by spdk_nvme_ctrlr_get_num_ns() may also change.
+ *
+ * \param ctrlr Opaque handle to NVMe controller.
+ *
+ * \return 0 on success, -1 on failure, -ENOTSUP if subsystem reset is not supported.
+ */
+int spdk_nvme_ctrlr_reset_subsystem(struct spdk_nvme_ctrlr *ctrlr);
+
 /**
  * Fail the given NVMe controller.
  *

lib/nvme/nvme_ctrlr.c

@@ -91,6 +91,13 @@ nvme_ctrlr_get_cmbsz(struct spdk_nvme_ctrlr *ctrlr, union spdk_nvme_cmbsz_regist
 					      &cmbsz->raw);
 }
 
+static int
+nvme_ctrlr_set_nssr(struct spdk_nvme_ctrlr *ctrlr, uint32_t nssr_value)
+{
+	return nvme_transport_ctrlr_set_reg_4(ctrlr, offsetof(struct spdk_nvme_registers, nssr),
+					      nssr_value);
+}
+
 bool
 nvme_ctrlr_multi_iocs_enabled(struct spdk_nvme_ctrlr *ctrlr)
 {
@@ -1431,6 +1438,32 @@ out:
 	return rc;
 }
 
+int
+spdk_nvme_ctrlr_reset_subsystem(struct spdk_nvme_ctrlr *ctrlr)
+{
+	union spdk_nvme_cap_register cap;
+	int rc = 0;
+
+	cap = spdk_nvme_ctrlr_get_regs_cap(ctrlr);
+	if (cap.bits.nssrs == 0) {
+		SPDK_WARNLOG("subsystem reset is not supported\n");
+		return -ENOTSUP;
+	}
+
+	SPDK_NOTICELOG("resetting subsystem\n");
+	nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
+	ctrlr->is_resetting = true;
+	rc = nvme_ctrlr_set_nssr(ctrlr, SPDK_NVME_NSSR_VALUE);
+	ctrlr->is_resetting = false;
+
+	nvme_robust_mutex_unlock(&ctrlr->ctrlr_lock);
+	/*
+	 * No more cleanup at this point like in the ctrlr reset. A subsystem reset will cause
+	 * a hot remove for PCIe transport. The hot remove handling does all the necessary ctrlr cleanup.
+	 */
+	return rc;
+}
+
 int
 spdk_nvme_ctrlr_set_trid(struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_transport_id *trid)
 {

lib/nvme/spdk_nvme.map

@@ -30,6 +30,7 @@
 	spdk_nvme_ctrlr_is_discovery;
 	spdk_nvme_ctrlr_get_default_ctrlr_opts;
 	spdk_nvme_ctrlr_set_trid;
+	spdk_nvme_ctrlr_reset_subsystem;
 	spdk_nvme_ctrlr_reset;
 	spdk_nvme_ctrlr_fail;
 	spdk_nvme_ctrlr_is_failed;